Watch the PathMap™ discovery engine take a list of keywords in AGI mode in a 5 pass run utilizing the patent pending invention I made and I want to share with the world. The run-set in the video took under 15 minutes on my MetroPCS phone using Google Gemini Free tier.
It is super RAG optimized anyway and you can use cheap models for veridical builds and then use "smarter models" for evaluation. Demo coming soon! At the end of the video I challenge Google Search engine and the software wins. Many AI will reject new discoveries, just watch...
Built for unparalleled (as far as I know) efficiency. Pathmap architects a unified dataset and renders data-veridical visualizations in minutes—entirely even on a 4GB budget mobile browser constraint!
It is local-first but I envision many embodiments that could use this tech! (again, a friendly reminder, I am patenting this so that it can be properly managed and contemplated evaluated and enforced).
See what happens when search AI falsely accuses my pipeline of hallucinating. I pushed back because AI often rejects new truths saying the PMID are hallucinated because they are too new... the workbench uses pubmed API and are real...
or it will say the software conflated something, as you will see in the video above, based on a lack of evidence and thus making it seem like the data I generated is trash, which it is not. I have studied ALS since last June and so I am able in the video to push back and show the Google AI its erroneous way.
The software makes real discoveries, but they need real human peer review. I am not a medical person, but I studied ALS and published a paper to then to my horror find it riddled with AI confabulations.
In June 2025, I was inspired by Eric Dane's ALS interview. I spent months using AI chatbots to research and published 'The Drusen Switch' in early January of 2026. Three days later I learned about thermodynamic affinity and that BMAA cannot chelate Zn2 from SOD1, and that the citations were riddled with hallucinations. THAT'S why I made PathMap (and because I am mostly broke but was inspired by Eric Dane's ALS interview last June, well, I made it so I can use it for free if ever I cannot afford the ungodly internet prices so there ya have the gist of the origin story.)
When challenged with an anatomical impossibility (Retina to Spinal Cord projection), I used my own research and human literature based discoveries to rebutt the AI with physiological plausibility.
Not all features will be present in release form initially... this is for ALPHA TESTERS and bio-hackers and will be available in demo form and full form soon, NON-COMMERCIAL USE (when I can afford legal help I will release freelancer and institutional licenses)... a great vacuum in many areas such as paralegals, writers, judges, teachers, researchers, technical writers, accountants, and any veridcal-to-data service involved industry... on the other hand, instead getting hallicinated data and spinning our wheels and waging unneeded wars, we can perhaps find a better use for this most certainly ARTIFICIAL "intelligence"...).
Bring your own API key. Natively interfaces with Google Gemini, OpenAI, OpenRouter. Clinical/institutional license will have an airgap-capable build I designed.
The Universal AI Workbench generates complex, modular analytics offline. Instantly renders Logic Gate Flowcharts, Contradiction Topologies, Semantic Attractor Networks, and Gap Distributions utilizing Mermaid.js.
Features "Auto-Explore" for hands-free hypothesis generation and a "Smart Follow-Up" AGI (precursory) agent that analyzes the print-menu analytics to automatically suggest your next logical research step.
Need specific lab metrics? The new Datapoint Architect lets you inject dynamic JSON extraction rules mid-pipeline, instantly generating isolated visual dashboards for your custom metrics. The AGI also creates its own datapoints which then spawn reports which then are able to be viewed by the AGI as it recursively builds its dataset for discovery.
Select your language and the engine utilizes an AI call to translate the entire application UI and core system prompts without breaking system syntax, into YOUR language!
Generate thousands of veridical-to-source quotes using the pass/fail guaranteed verbatim enforcement, which allows for better results and more accurate use of grounding
A note on that, remember, AI can still make logic-based mistakes.
For example, smaller LLMs are great "harvesters" of data but they may struggle to calculate efficiently across the perfect dataset created. This can show up in many ways, and I have seen it most often when evolving evidence emerges and the AI gives preference to longer established data, and disregards a new discovery. Adding the Swanson's datapoint helped a lot with this.
Merge local JSON traces (paid version... a lot of demo features are nerfed because it is overwhelming especially now that I am finishing the V11.7 with gated semantic drift and veridical metrics and veridical enforcement... it is a workbench but it is also a hell of a researcher). Print a beautifully formatted synthesis with perfect APA citations.
Pathmap v11.0 is currently in private Alpha. Join the waitlist to be notified the moment slots open up.
Sign Up For Launch NotificationsRESEARCH DISCLOSURE: Pathmap is a Literature-Based Discovery engine and precursor to AGI, offering autonomous research, while fitting in your pocket (runs on a cell phone or computer).
Results do not constitute medical, legal, or professional advice.
All logical assertions, generated outputs, and hypothesized pathways must be independently verified by human professionals.
© 2026 Artificial General Intelligence LLC. All Rights Reserved.
Grand Rapids, MI 49508
The Artificial General Intelligence™ or AGI™ claims made on this page are validated by the fact that veridical AI is an absolute foundational pre-requisite for AGI, and from here, it is just a matter of building the brain, which I plan to do at mindbuilder(.ai) someday.
PATENT PENDING
♥Joshua